/*
* Release blocks associated with the inode ip and stored in the indirect
* block bn. Blocks are free'd in LIFO order up to (but not including)
* lastbn. If level is greater than SINGLE, the block is an indirect block
* and recursive calls to indirtrunc must be used to cleanse other indirect
* blocks.
*
* NB: triple indirect blocks are untested.
*/
static int
ext2fs_indirtrunc(struct inode *ip, int32_t lbn, int32_t dbn, int32_t lastbn, int level, long *countp)
{
int i;
struct buf *bp;
struct m_ext2fs *fs = ip->i_e2fs;
int32_t *bap;
struct vnode *vp;
int32_t *copy = NULL, nb, nlbn, last;
long blkcount, factor;
int nblocks, blocksreleased = 0;
int error = 0, allerror = 0;
/*
* Calculate index in current block of last
* block to be kept. -1 indicates the entire
* block so we need not calculate the index.
*/
factor = 1;
for (i = SINGLE; i < level; i++)
factor *= NINDIR(fs);
last = lastbn;
if (lastbn > 0)
last /= factor;
nblocks = btodb(fs->e2fs_bsize);
/*
* Get buffer of block pointers, zero those entries corresponding
* to blocks to be free'd, and update on disk copy first. Since
* double(triple) indirect before single(double) indirect, calls
* to bmap on these blocks will fail. However, we already have
* the on disk address, so we have to set the b_blkno field
* explicitly instead of letting bread do everything for us.
*/
vp = ITOV(ip);
bp = getblk(vp, lbn, (int)fs->e2fs_bsize, 0, 0);
if (!(bp->b_flags & (B_DONE | B_DELWRI))) {
curproc->p_ru.ru_inblock++; /* pay for read */
bcstats.pendingreads++;
bcstats.numreads++;
bp->b_flags |= B_READ;
if (bp->b_bcount > bp->b_bufsize)
panic("ext2fs_indirtrunc: bad buffer size");
bp->b_blkno = dbn;
VOP_STRATEGY(bp);
error = biowait(bp);
}
if (error) {
brelse(bp);
*countp = 0;
return (error);
}
bap = (int32_t *)bp->b_data;
if (lastbn >= 0) {
copy = malloc(fs->e2fs_bsize, M_TEMP, M_WAITOK);
memcpy(copy, bap, fs->e2fs_bsize);
memset(&bap[last + 1], 0,
(NINDIR(fs) - (last + 1)) * sizeof(u_int32_t));
error = bwrite(bp);
if (error)
allerror = error;
bap = copy;
}
/*
* Recursively free totally unused blocks.
*/
for (i = NINDIR(fs) - 1,
nlbn = lbn + 1 - i * factor; i > last;
i--, nlbn += factor) {
nb = letoh32(bap[i]);
if (nb == 0)
continue;
if (level > SINGLE) {
error = ext2fs_indirtrunc(ip, nlbn, fsbtodb(fs, nb),
(int32_t)-1, level - 1,
&blkcount);
if (error)
allerror = error;
blocksreleased += blkcount;
}
ext2fs_blkfree(ip, nb);
blocksreleased += nblocks;
}
/*
* Recursively free last partial block.
*/
if (level > SINGLE && lastbn >= 0) {
last = lastbn % factor;
nb = letoh32(bap[i]);
if (nb != 0) {
error = ext2fs_indirtrunc(ip, nlbn, fsbtodb(fs, nb),
last, level - 1, &blkcount);
if (error)
allerror = error;
blocksreleased += blkcount;
}
}
if (copy != NULL) {
free(copy, M_TEMP, fs->e2fs_bsize);
} else {
bp->b_flags |= B_INVAL;
brelse(bp);
}
*countp = blocksreleased;
return (allerror);
}
/*
* Balloc defines the structure of file system storage
* by allocating the physical blocks on a device given
* the inode and the logical block number in a file.
*/
int
ext2fs_balloc(struct inode *ip, daddr_t bn, int size,
kauth_cred_t cred, struct buf **bpp, int flags)
{
struct m_ext2fs *fs;
daddr_t nb;
struct buf *bp, *nbp;
struct vnode *vp = ITOV(ip);
struct indir indirs[EXT2FS_NIADDR + 2];
daddr_t newb, lbn, pref;
int32_t *bap; /* XXX ondisk32 */
int num, i, error;
u_int deallocated;
daddr_t *blkp, *allocblk, allociblk[EXT2FS_NIADDR + 1];
int32_t *allocib; /* XXX ondisk32 */
int unwindidx = -1;
UVMHIST_FUNC("ext2fs_balloc"); UVMHIST_CALLED(ubchist);
UVMHIST_LOG(ubchist, "bn 0x%x", bn,0,0,0);
if (bpp != NULL) {
*bpp = NULL;
}
if (bn < 0)
return (EFBIG);
fs = ip->i_e2fs;
lbn = bn;
/*
* The first EXT2FS_NDADDR blocks are direct blocks
*/
if (bn < EXT2FS_NDADDR) {
/* XXX ondisk32 */
nb = fs2h32(ip->i_e2fs_blocks[bn]);
if (nb != 0) {
/*
* the block is already allocated, just read it.
*/
if (bpp != NULL) {
error = bread(vp, bn, fs->e2fs_bsize, NOCRED,
B_MODIFY, &bp);
if (error) {
return (error);
}
*bpp = bp;
}
return (0);
}
/*
* allocate a new direct block.
*/
error = ext2fs_alloc(ip, bn,
ext2fs_blkpref(ip, bn, bn, &ip->i_e2fs_blocks[0]),
cred, &newb);
if (error)
return (error);
ip->i_e2fs_last_lblk = lbn;
ip->i_e2fs_last_blk = newb;
/* XXX ondisk32 */
ip->i_e2fs_blocks[bn] = h2fs32((int32_t)newb);
ip->i_flag |= IN_CHANGE | IN_UPDATE;
if (bpp != NULL) {
bp = getblk(vp, bn, fs->e2fs_bsize, 0, 0);
bp->b_blkno = EXT2_FSBTODB(fs, newb);
if (flags & B_CLRBUF)
clrbuf(bp);
*bpp = bp;
}
return (0);
}
/*
* Determine the number of levels of indirection.
*/
pref = 0;
if ((error = ufs_getlbns(vp, bn, indirs, &num)) != 0)
return(error);
#ifdef DIAGNOSTIC
if (num < 1)
panic ("ext2fs_balloc: ufs_getlbns returned indirect block\n");
#endif
/*
* Fetch the first indirect block allocating if necessary.
*/
--num;
/* XXX ondisk32 */
nb = fs2h32(ip->i_e2fs_blocks[EXT2FS_NDADDR + indirs[0].in_off]);
allocib = NULL;
allocblk = allociblk;
if (nb == 0) {
pref = ext2fs_blkpref(ip, lbn, 0, (int32_t *)0);
error = ext2fs_alloc(ip, lbn, pref, cred, &newb);
if (error)
return (error);
nb = newb;
*allocblk++ = nb;
ip->i_e2fs_last_blk = newb;
bp = getblk(vp, indirs[1].in_lbn, fs->e2fs_bsize, 0, 0);
bp->b_blkno = EXT2_FSBTODB(fs, newb);
clrbuf(bp);
/*
* Write synchronously so that indirect blocks
* never point at garbage.
*/
if ((error = bwrite(bp)) != 0)
goto fail;
unwindidx = 0;
allocib = &ip->i_e2fs_blocks[EXT2FS_NDADDR + indirs[0].in_off];
/* XXX ondisk32 */
*allocib = h2fs32((int32_t)newb);
ip->i_flag |= IN_CHANGE | IN_UPDATE;
}
/*
* Fetch through the indirect blocks, allocating as necessary.
*/
for (i = 1;;) {
error = bread(vp,
indirs[i].in_lbn, (int)fs->e2fs_bsize, NOCRED, 0, &bp);
if (error) {
goto fail;
}
bap = (int32_t *)bp->b_data; /* XXX ondisk32 */
nb = fs2h32(bap[indirs[i].in_off]);
if (i == num)
break;
i++;
if (nb != 0) {
brelse(bp, 0);
continue;
}
pref = ext2fs_blkpref(ip, lbn, 0, (int32_t *)0);
error = ext2fs_alloc(ip, lbn, pref, cred, &newb);
if (error) {
brelse(bp, 0);
goto fail;
}
nb = newb;
*allocblk++ = nb;
ip->i_e2fs_last_blk = newb;
nbp = getblk(vp, indirs[i].in_lbn, fs->e2fs_bsize, 0, 0);
nbp->b_blkno = EXT2_FSBTODB(fs, nb);
clrbuf(nbp);
/*
* Write synchronously so that indirect blocks
* never point at garbage.
*/
if ((error = bwrite(nbp)) != 0) {
brelse(bp, 0);
goto fail;
}
if (unwindidx < 0)
unwindidx = i - 1;
/* XXX ondisk32 */
bap[indirs[i - 1].in_off] = h2fs32((int32_t)nb);
/*
* If required, write synchronously, otherwise use
* delayed write.
*/
if (flags & B_SYNC) {
bwrite(bp);
} else {
bdwrite(bp);
}
}
/*
* Get the data block, allocating if necessary.
*/
if (nb == 0) {
pref = ext2fs_blkpref(ip, lbn, indirs[num].in_off, &bap[0]);
error = ext2fs_alloc(ip, lbn, pref, cred, &newb);
if (error) {
brelse(bp, 0);
goto fail;
}
nb = newb;
*allocblk++ = nb;
ip->i_e2fs_last_lblk = lbn;
ip->i_e2fs_last_blk = newb;
/* XXX ondisk32 */
bap[indirs[num].in_off] = h2fs32((int32_t)nb);
/*
* If required, write synchronously, otherwise use
* delayed write.
*/
if (flags & B_SYNC) {
bwrite(bp);
} else {
bdwrite(bp);
}
if (bpp != NULL) {
nbp = getblk(vp, lbn, fs->e2fs_bsize, 0, 0);
nbp->b_blkno = EXT2_FSBTODB(fs, nb);
if (flags & B_CLRBUF)
clrbuf(nbp);
*bpp = nbp;
}
return (0);
}
brelse(bp, 0);
if (bpp != NULL) {
if (flags & B_CLRBUF) {
error = bread(vp, lbn, (int)fs->e2fs_bsize, NOCRED,
B_MODIFY, &nbp);
if (error) {
goto fail;
}
} else {
nbp = getblk(vp, lbn, fs->e2fs_bsize, 0, 0);
nbp->b_blkno = EXT2_FSBTODB(fs, nb);
}
*bpp = nbp;
}
return (0);
fail:
/*
* If we have failed part way through block allocation, we
* have to deallocate any indirect blocks that we have allocated.
*/
for (deallocated = 0, blkp = allociblk; blkp < allocblk; blkp++) {
ext2fs_blkfree(ip, *blkp);
deallocated += fs->e2fs_bsize;
}
if (unwindidx >= 0) {
if (unwindidx == 0) {
*allocib = 0;
} else {
int r;
r = bread(vp, indirs[unwindidx].in_lbn,
(int)fs->e2fs_bsize, NOCRED, B_MODIFY, &bp);
if (r) {
panic("Could not unwind indirect block, error %d", r);
} else {
bap = (int32_t *)bp->b_data; /* XXX ondisk32 */
bap[indirs[unwindidx].in_off] = 0;
if (flags & B_SYNC)
bwrite(bp);
else
bdwrite(bp);
}
}
for (i = unwindidx + 1; i <= num; i++) {
bp = getblk(vp, indirs[i].in_lbn, (int)fs->e2fs_bsize,
0, 0);
brelse(bp, BC_INVAL);
}
}
if (deallocated) {
ext2fs_setnblock(ip, ext2fs_nblock(ip) - btodb(deallocated));
ip->i_e2fs_flags |= IN_CHANGE | IN_UPDATE;
}
return error;
}
/*
* Truncate the inode oip to at most length size, freeing the
* disk blocks.
*/
int
ext2fs_truncate(struct inode *oip, off_t length, int flags, struct ucred *cred)
{
struct vnode *ovp = ITOV(oip);
int32_t lastblock;
int32_t bn, lbn, lastiblock[NIADDR], indir_lbn[NIADDR];
int32_t oldblks[NDADDR + NIADDR], newblks[NDADDR + NIADDR];
struct m_ext2fs *fs;
struct buf *bp;
int offset, size, level;
long count, nblocks, vflags, blocksreleased = 0;
int i;
int aflags, error, allerror;
off_t osize;
if (length < 0)
return (EINVAL);
if (ovp->v_type != VREG &&
ovp->v_type != VDIR &&
ovp->v_type != VLNK)
return (0);
if (ovp->v_type == VLNK && ext2fs_size(oip) < EXT2_MAXSYMLINKLEN) {
#ifdef DIAGNOSTIC
if (length != 0)
panic("ext2fs_truncate: partial truncate of symlink");
#endif
memset(&oip->i_e2din->e2di_shortlink, 0, ext2fs_size(oip));
(void)ext2fs_setsize(oip, 0);
oip->i_flag |= IN_CHANGE | IN_UPDATE;
return (ext2fs_update(oip, 1));
}
if (ext2fs_size(oip) == length) {
oip->i_flag |= IN_CHANGE | IN_UPDATE;
return (ext2fs_update(oip, 0));
}
fs = oip->i_e2fs;
osize = ext2fs_size(oip);
/*
* Lengthen the size of the file. We must ensure that the
* last byte of the file is allocated. Since the smallest
* value of osize is 0, length will be at least 1.
*/
if (osize < length) {
#if 0 /* XXX */
if (length > fs->fs_maxfilesize)
return (EFBIG);
#endif
offset = blkoff(fs, length - 1);
lbn = lblkno(fs, length - 1);
aflags = B_CLRBUF;
if (flags & IO_SYNC)
aflags |= B_SYNC;
error = ext2fs_buf_alloc(oip, lbn, offset + 1, cred, &bp,
aflags);
if (error)
return (error);
(void)ext2fs_setsize(oip, length);
uvm_vnp_setsize(ovp, length);
uvm_vnp_uncache(ovp);
if (aflags & B_SYNC)
bwrite(bp);
else
bawrite(bp);
oip->i_flag |= IN_CHANGE | IN_UPDATE;
return (ext2fs_update(oip, 1));
}
/*
* Shorten the size of the file. If the file is not being
* truncated to a block boundry, the contents of the
* partial block following the end of the file must be
* zero'ed in case it ever become accessible again because
* of subsequent file growth.
*/
offset = blkoff(fs, length);
if (offset == 0) {
(void)ext2fs_setsize(oip, length);
} else {
lbn = lblkno(fs, length);
aflags = B_CLRBUF;
if (flags & IO_SYNC)
aflags |= B_SYNC;
error = ext2fs_buf_alloc(oip, lbn, offset, cred, &bp,
aflags);
if (error)
return (error);
(void)ext2fs_setsize(oip, length);
size = fs->e2fs_bsize;
uvm_vnp_setsize(ovp, length);
uvm_vnp_uncache(ovp);
memset(bp->b_data + offset, 0, size - offset);
bp->b_bcount = size;
if (aflags & B_SYNC)
bwrite(bp);
else
bawrite(bp);
}
/*
* Calculate index into inode's block list of
* last direct and indirect blocks (if any)
* which we want to keep. Lastblock is -1 when
* the file is truncated to 0.
*/
lastblock = lblkno(fs, length + fs->e2fs_bsize - 1) - 1;
lastiblock[SINGLE] = lastblock - NDADDR;
lastiblock[DOUBLE] = lastiblock[SINGLE] - NINDIR(fs);
lastiblock[TRIPLE] = lastiblock[DOUBLE] - NINDIR(fs) * NINDIR(fs);
nblocks = btodb(fs->e2fs_bsize);
/*
* Update file and block pointers on disk before we start freeing
* blocks. If we crash before free'ing blocks below, the blocks
* will be returned to the free list. lastiblock values are also
* normalized to -1 for calls to ext2fs_indirtrunc below.
*/
memcpy(oldblks, &oip->i_e2fs_blocks[0], sizeof(oldblks));
for (level = TRIPLE; level >= SINGLE; level--)
if (lastiblock[level] < 0) {
oip->i_e2fs_blocks[NDADDR + level] = 0;
lastiblock[level] = -1;
}
for (i = NDADDR - 1; i > lastblock; i--)
oip->i_e2fs_blocks[i] = 0;
oip->i_flag |= IN_CHANGE | IN_UPDATE;
if ((error = ext2fs_update(oip, 1)) != 0)
allerror = error;
/*
* Having written the new inode to disk, save its new configuration
* and put back the old block pointers long enough to process them.
* Note that we save the new block configuration so we can check it
* when we are done.
*/
memcpy(newblks, &oip->i_e2fs_blocks[0], sizeof(newblks));
memcpy(&oip->i_e2fs_blocks[0], oldblks, sizeof(oldblks));
(void)ext2fs_setsize(oip, osize);
vflags = ((length > 0) ? V_SAVE : 0) | V_SAVEMETA;
allerror = vinvalbuf(ovp, vflags, cred, curproc, 0, 0);
/*
* Indirect blocks first.
*/
indir_lbn[SINGLE] = -NDADDR;
indir_lbn[DOUBLE] = indir_lbn[SINGLE] - NINDIR(fs) -1;
indir_lbn[TRIPLE] = indir_lbn[DOUBLE] - NINDIR(fs) * NINDIR(fs) - 1;
for (level = TRIPLE; level >= SINGLE; level--) {
bn = letoh32(oip->i_e2fs_blocks[NDADDR + level]);
if (bn != 0) {
error = ext2fs_indirtrunc(oip, indir_lbn[level],
fsbtodb(fs, bn), lastiblock[level], level, &count);
if (error)
allerror = error;
blocksreleased += count;
if (lastiblock[level] < 0) {
oip->i_e2fs_blocks[NDADDR + level] = 0;
ext2fs_blkfree(oip, bn);
blocksreleased += nblocks;
}
}
if (lastiblock[level] >= 0)
goto done;
}
/*
* All whole direct blocks or frags.
*/
for (i = NDADDR - 1; i > lastblock; i--) {
bn = letoh32(oip->i_e2fs_blocks[i]);
if (bn == 0)
continue;
oip->i_e2fs_blocks[i] = 0;
ext2fs_blkfree(oip, bn);
blocksreleased += btodb(fs->e2fs_bsize);
}
done:
#ifdef DIAGNOSTIC
for (level = SINGLE; level <= TRIPLE; level++)
if (newblks[NDADDR + level] !=
oip->i_e2fs_blocks[NDADDR + level])
panic("ext2fs_truncate1");
for (i = 0; i < NDADDR; i++)
if (newblks[i] != oip->i_e2fs_blocks[i])
panic("ext2fs_truncate2");
if (length == 0 &&
(!LIST_EMPTY(&ovp->v_cleanblkhd) ||
!LIST_EMPTY(&ovp->v_dirtyblkhd)))
panic("ext2fs_truncate3");
#endif /* DIAGNOSTIC */
/*
* Put back the real size.
*/
(void)ext2fs_setsize(oip, length);
if (blocksreleased >= oip->i_e2fs_nblock)
oip->i_e2fs_nblock = 0;
else
oip->i_e2fs_nblock -= blocksreleased;
oip->i_flag |= IN_CHANGE;
return (allerror);
}